NVIDIA DROPS RTX SPARK SUPERCHIP AND COSMOS 3 AT COMPUTEX — A LAPTOP THAT RUNS AI AGENTS AND A MODEL THAT GIVES ROBOTS A BRAIN Alongside the Vera Rubin enterprise announcement, Jensen Huang unveiled two products at GTC Taipei that push powerful AI far beyond the data center. The RTX Spark superchip combines a 20-core Arm CPU with a Blackwell GPU and 128 gigabytes of unified memory into a single package designed to run AI agents locally on Windows laptops and compact desktops. ASUS, Dell, HP, Lenovo, Microsoft Surface, and MSI will ship RTX Spark machines this fall. NVIDIA says it puts one petaflop of AI compute into a device you can carry. The other announcement is larger in scope. Cosmos 3 is NVIDIA’s frontier foundation model for physical AI, built on a mixture-of-transformers architecture that functions simultaneously as a visual language model, a world model, a simulator, and a foundation for robot action policies. Huang said Cosmos 3 can simulate physical environments and generate training data for robots and autonomous vehicles at speeds that would take physical testing months to replicate. It is available now via Hugging Face and NVIDIA’s NIM microservices. The move to run serious AI locally changes the privacy and latency equation for businesses and developers who cannot or will not pipe sensitive data through cloud servers. This is NVIDIA making a play for every layer of the stack at once — data centers with Vera Rubin, consumers and enterprises with RTX Spark, and robot builders with Cosmos 3. Keywords: NVIDIA RTX Spark, Cosmos 3 AI model, physical AI 2026, AI PC chip